Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

The Representational Alignment between Humans and Language Models is implicitly driven by a Concreteness Effect

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper uses behavioral similarity judgments and embedding ablations to argue that the apparent alignment between humans and language models on single-word meaning is largely an alignment on the concreteness dimension.

desk verdict Useful conditional evidence that concreteness is a shared axis of human-model alignment for single nouns; the ablation-based causal claim needs stronger statistics and a linearity check before it will convince. read the letter →

arxiv 2505.15682 v1 pith:QFF2SCJY submitted 2025-05-21 cs.CL

Cosimo Iaia , Bhavin Choksi , Emily Wiebers , Gemma Roig , Christian J. Fiebach This is my paper
classification cs.CL
keywords concretenessrepresentationalsimilarityanalysislanguagemodelswordembeddingsodd-one-outtaskhuman-modelalignmentsemanticrepresentationablationstudy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Using behavioral odd-one-out judgments for 40 German nouns, the paper builds an implicit human similarity space and compares it with embeddings from five language models. It finds significant alignment between the two, and shows that both are aligned to an explicit concreteness-rating space. When concreteness is removed from the model embeddings by linear regression, the human-model alignment drops by 20.6% on average, significantly for every model, while removing word frequency, word length, or orthographic similarity changes it by less than 7.6%. The paper concludes that the apparent agreement between humans and language models about word meaning is substantially carried by the concrete-versus-abstract dimension, not by other tested lexical features.

What carries the argument

The machinery is Representational Similarity Analysis (RSA), which turns each space, whether human odd-one-out choices, model embeddings, or feature ratings, into a representational dissimilarity matrix (RDM) of pairwise distances and compares RDMs by Spearman correlation. The causal work is done by the ablation procedure: a Ridge regression predicts each model's word embeddings from a feature rating such as concreteness, and the residuals are treated as embeddings with that feature removed; re-running RSA on the residual RDMs quantifies how much of the human-model alignment depended on that feature. The concreteness ratings used for stimulus selection and ablation come from an automatic German norm set, validated against participants' own ratings (r = 0.86).

What would settle it

Train a nonlinear decoder, such as a small multilayer network, on the ridge-residual embeddings to predict participants' concreteness ratings; if it predicts held-out words above chance, the residuals still carry concreteness-related structure, and the linear-ablation drop cannot be read as the full contribution of concreteness to human-model alignment.

Watch

Extended reading notes

Core claim

The core claim is that concreteness is the principal shared dimension organizing single-word semantic representations in humans and in language models, even though neither system was explicitly trained or probed on it. According to the paper, the representational dissimilarity matrix derived from participants' odd-one-out choices correlates significantly with embedding-based RDMs for fastText, word2vec, BERT base, BERT large, and GPT2; both the human and model spaces correlate with a concreteness-rating RDM; and linear ablation of concreteness from the embeddings reduces the human-model RSA correlation by an average of 20.6% across models, with every model showing a significant drop, whereas ablation of frequency, length, or orthographic similarity produces effects below 7.6%. The authors take this as evidence that humans and language models converge on the concreteness dimension but not on other dimensions.

Load-bearing premise

The conclusion assumes concreteness is carried linearly in word embeddings, so regressing it out removes it; if concreteness is also encoded nonlinearly, the residual embeddings still contain concreteness-like structure and the measured drop does not isolate concreteness.

Editorial extensions

If this is right

  • Word-similarity benchmarks that compare model embeddings with human judgments will need to control for concreteness before claiming broader semantic agreement.
  • For model evaluation, matching human odd-one-out behavior on single nouns may be achievable mainly by reproducing the concrete/abstract split, so passing such benchmarks is weak evidence of deep semantic alignment.
  • For psychology, the results show that an implicit concreteness axis can be recovered from similarity judgments without ever asking participants about concreteness.
  • For engineering, improving human-model alignment beyond concreteness would require training objectives that capture other semantic structure rather than relying on current distributional statistics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: Because the ablation removes only the linearly decodable part of concreteness, the 20.6% drop should be read as the linear component of concreteness's contribution; a nonlinear probe could show the true contribution is larger.
  • Editorial inference: The stimulus set was deliberately built to maximize concreteness variance, so the human RDM is partly structured around that axis; replication with words matched on concreteness but varying other dimensions could reveal alignment that this design underweights.
  • Editorial inference: A natural extension is to test contextualized models with the same word in concrete and abstract senses; if the concreteness account is right, model-human alignment should move with the sense shift within a single word type.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper investigates whether the representational alignment between human similarity judgments and language model embeddings for single words is driven by the concreteness of word meanings. The authors collected odd-one-out judgments on 40 carefully selected German nouns, built a behavioral representational dissimilarity matrix (RDM), and compared it via RSA to embeddings from fastText, word2vec, BERT base, BERT large, and GPT2. They report significant human-model alignment, and that both spaces are independently aligned to explicit concreteness ratings. Using a linear residualization ablation (Ridge regression), they find that removing concreteness from model embeddings reduces human-model alignment by an average of 20.6%, while removing frequency, length, or orthographic similarity (OLD20) has effects below 7.6%. They conclude that concreteness critically drives the human-model alignment.

Significance. If the result holds, it would provide evidence that a concreteness dimension is shared between human semantic representations and LLM embeddings, with implications for cognitive modeling, interpretability, and the evaluation of semantic representations. The paper's strengths include the use of an implicit behavioral task, multiple model architectures, and a comparison against several psycholinguistic control features. The authors also commit to releasing code and data, which supports reproducibility. However, the central claim rests on a linearity assumption that is not validated, a stimulus set that maximizes concreteness variance while restricting control-feature variance, and significance tests that treat non-independent RDM entries as independent. These issues make the headline conclusion stronger than the current evidence supports.

major comments (5)
  1. [3.6] The ablation conclusion assumes that a linear Ridge regression fully removes the targeted feature from the embeddings and that this removal is comparable across features. The manuscript states that the approach 'assumes a linear relationship' but provides no evidence that this assumption holds equally for concreteness and for word frequency, word length, and OLD20. If concreteness is encoded in a strong linear subspace while the control features are encoded more nonlinearly, linear residualization will preferentially delete concreteness structure and produce the observed asymmetry even if the true contributions to alignment are similar. Please report the R² of the Ridge regression for each feature and model, and validate the residualization with a nonlinear removal method or a decoding analysis on the residuals.
  2. [3.2 and 4.3] The stimulus set was selected from the cluster with the highest concreteness variance, with concrete and abstract words selected beyond one standard deviation, while control features (frequency, length, OLD20) were matched within one standard deviation. This design guarantees that concreteness dominates the RDM and restricts the variance of the control features. Consequently, the small ablation effects for frequency, length, and OLD20 may simply reflect their limited range in the stimulus set rather than their irrelevance to human-model alignment. The claim that alignment is 'not by other dimensions' is too strong. Please quantify the variances of all features in the chosen set and, where possible, provide positive controls demonstrating that the control features are detectable in the model RDMs under the same ablation procedure.
  3. [4.3 and Table 1] The results for GPT2 are internally inconsistent. Table 1 shows no significant alignment between GPT2 and either the explicit or automatic concreteness RDM (ρ = .06 and .01, both n.s.), yet Section 4.3 reports that removing concreteness significantly reduces GPT2's alignment with the behavioral RDM (p < .001). If GPT2 does not represent the concreteness ratings linearly, the ablation should have little effect on its RDM. The authors should explain this discrepancy or temper the claim that the concreteness effect is consistent across all models.
  4. [4.3] The significance of RSA correlations and ablation drops is assessed with the Williams test on correlations between RDMs that contain 780 off-diagonal entries derived from only 40 words. These entries are not independent, so the effective sample size is much smaller than 780, and the p-values are likely anti-conservative. Please re-evaluate the significance of the alignment and the ablation effects using item-level permutation tests (e.g., permuting word labels) or bootstrap over participants and words, and report the resulting confidence intervals.
  5. [Appendix A] The imageability control is not fully conclusive for the uniqueness claim. Imageability correlates .93 with concreteness in the stimulus set, and removing imageability also yields significant alignment drops for all models except word2vec. The conclusion that concreteness 'still plays a more critical role' is based on the relative size of the drop, but no statistical test is reported comparing the concreteness drop to the imageability drop. A direct comparison (e.g., bootstrap of the difference between drops) is needed to support the claim that concreteness is uniquely important.
minor comments (5)
  1. [3.5] The sentence 'we take the coded ratings (38 per combination)' is unclear: please specify how many participants contributed to each triplet and how the total of 9880 unique triplets was distributed across the 40 participants.
  2. [Figure 3] Please describe what the error bars in Figure 3 represent (e.g., standard error across participants, permutation intervals) and how the Williams test was applied to each model.
  3. [3.4] The reference to Müller (2015) is incomplete; please provide a full citation or a URL for the German word2vec embeddings.
  4. [Table 1] The note that 'behaviorally explicit concreteness ratings seem to perform better' than automatic ratings is not supported by a statistical comparison; either add a test or rephrase as a descriptive observation.
  5. [3.6] Please report the range of alpha values considered and the selected alpha for each Ridge regression, as well as the number of training embeddings used for each model (771 vs. 716).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the human-model alignment and concreteness-ablation results are empirical and not reducible to input definitions.

full rationale

The paper's derivation chain is self-contained at the level of definitions: the human implicit space is an odd-one-out RDM, the model spaces are cosine-distance RDMs from pretrained embeddings, and the explicit concreteness space is a rating-based RDM. The target quantity (the human-model RSA correlation and its change after ablation) is not used to define the concreteness feature or any control feature, and no parameter fitted to the outcome is relabeled as a prediction. The ridge residualization is explicitly an assumed linear operation ('this approach to removing features from embeddings assumes a linear relationship between the feature variables and the embedding'); this assumption limits the inference but does not make the observed 20.6% versus <7.6% drop a logical consequence of the input ratings, because the drop also depends on the independent behavioral odd-one-out RDM and on each model's embedding geometry. Stimulus selection (Section 3.2) intentionally maximized concreteness variance while matching controls, which raises legitimate questions about generalizability, but selecting items by a feature is not the same as defining the alignment result in terms of that feature, and the paper's Limitations section acknowledges the restricted word and frequency range. The only author self-citation (Fiebach and Friederici, 2004) supports a background neuroimaging claim and is externally established, so it is not load-bearing. No circular step can be exhibited with the required specificity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on three design choices: a stimulus set engineered to maximize concreteness variance while matching control features (Section 3.2), a linear decodability assumption for the ablation (Section 3.6), and RSA statistics computed over 780 non-independent entries of a 40-item RDM (Section 3.5). Free parameters are the cluster count k=19 and the 1-SD selection thresholds. No new entities are introduced.

free parameters (3)
  • Number of clusters k in stimulus selection = 19
    k-means++ clustering in Section 3.2; called 'empirically determined' with no stated criterion.
  • Concreteness and frequency extreme-selection threshold = 1 standard deviation beyond the cluster mean
    Section 3.2 selects 8 concrete and 8 abstract words beyond 1 SD (mean 4.95, SD 1.51) and frequent and infrequent words beyond 1 SD (mean 2.83, SD 0.77); this hand-chosen threshold makes concreteness the dominant axis of the stimulus set.
  • Control-feature matching window = within 1 SD of the cluster mean
    Words were matched for frequency, length, and OLD20 within 1 SD (Section 3.2); the restricted variance of control features underpins the null results for those features.
assumptions (4)
  • domain assumption Spearman correlation between RDMs measures representational alignment
    RSA framework (Kriegeskorte et al. 2008), used in Sections 3.5 to 4.3; the validity of comparing geometries across spaces is taken from the RSA literature.
  • domain assumption Odd-one-out choices reflect pairwise similarity, coded as 1 for the two unchosen words and 0 for pairs involving the chosen word
    Section 3.5, following Turini and Võ (2022); fails if participants select the odd word based on a single-word property rather than pairwise comparisons.
  • ad hoc to paper Concreteness is linearly decodable from word embeddings, so Ridge residuals are concreteness-free
    Section 3.6; the authors state the linearity assumption. Nonlinear concreteness encoding would leave concreteness in the residuals and break the ablation conclusion.
  • domain assumption Token-averaged non-contextual BERT and GPT2 embeddings are adequate single-word representations
    Section 3.4; averaging byte-level tokens discards subword and positional information that may be relevant to concreteness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Representational Alignment between Humans and Language Models is implicitly driven by a Concreteness Effect." pith.science (2026). https://pith.science/paper/QFF2SCJY

@misc{pith2026250515682,
  author       = {Pith},
  title        = {Pith review of: The Representational Alignment between Humans and Language Models is implicitly driven by a Concreteness Effect},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QFF2SCJY}},
  note         = {Machine review of arXiv:2505.15682}
}
read the original abstract

The nouns of our language refer to either concrete entities (like a table) or abstract concepts (like justice or love), and cognitive psychology has established that concreteness influences how words are processed. Accordingly, understanding how concreteness is represented in our mind and brain is a central question in psychology, neuroscience, and computational linguistics. While the advent of powerful language models has allowed for quantitative inquiries into the nature of semantic representations, it remains largely underexplored how they represent concreteness. Here, we used behavioral judgments to estimate semantic distances implicitly used by humans, for a set of carefully selected abstract and concrete nouns. Using Representational Similarity Analysis, we find that the implicit representational space of participants and the semantic representations of language models are significantly aligned. We also find that both representational spaces are implicitly aligned to an explicit representation of concreteness, which was obtained from our participants using an additional concreteness rating task. Importantly, using ablation experiments, we demonstrate that the human-to-model alignment is substantially driven by concreteness, but not by other important word characteristics established in psycholinguistics. These results indicate that humans and language models converge on the concreteness dimension, but not on other dimensions.

Figures

Figures reproduced from arXiv: 2505.15682 by the authors.

Figure 1
Figure 1. Schematic of the approach: Words are sampled from the semantic space (top left) and used for an odd-one-out task. The English translations of the German words are Körper : Body, Gesicht : Face, Figur : Figure. For a set of 40 words, we collected a total of 9880 odd-one-out choices. These were converted into a representational dissimilarity matrix (RDM) reflecting pair-wise semantic distances between words. For each … view at source ↗
Figure 2
Figure 2. Partial correlations for the behavioral model (odd-one-out) and the computational models (language models): The representational space derived from the odd-on-out (in blue) is only correlated to the rated concreteness space, while language models (other colors) are aligned to other feature spaces as well. The representational spaces derived from all language models but GPT2 (in red) show alignment not only to concre… view at source ↗
Figure 3
Figure 3. Representational Similarity Analysis after removing each feature: Compared to the original correla￾tions between the non-ablated computational representation (lightest shade of blue) and the representation derived from the odd-one-out task, the biggest drop is observed when removing concreteness (dark blue) for all language models. (Williams’ test, *** p < .001, ** p < .01, * p < .05) on word concreteness. The impor… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Control Analysis: Representational Similarity analysis after removing further semantic features. The ablation approach reported in the main paper was repeated with further semantic dimensions, i.e., word imageability, word arousal, and word valence. Compared to the bas…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Large Language Models Show Signs of Alignment with Human Neurocognition During Abstract Reasoning

    q-bio.NC 2025-08 unverdicted novelty 6.0 of 10

    Only the largest tested LLMs (about 70 billion parameters) match human accuracy on an abstract reasoning task, and the internal geometry of their best layers correlates moderately with human frontal EEG activity.

Reference graph

Works this paper leans on

42 extracted references · 31 canonical work pages · cited by 1 Pith paper

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Jeffrey R Binder, Lisa L Conant, Colin J Humphries, Leonardo Fernandino, Stephen B Simons, Mario Aguilar, and Rutvik H Desai. 2016. Toward a brain-based componential semantic representation. Cognitive neuropsychology, 33(3-4):130--174

    work page 2016

  4. [4]

    Jeffrey R Binder, Chris F Westbury, Kristen A McKiernan, Edward T Possing, and David A Medler. 2005. Distinct brain systems for processing concrete and abstract concepts. Journal of cognitive neuroscience, 17(6):905--917

    work page 2005

  5. [5]

    Andrea Bruera, Yuan Tao, Andrew Anderson, Derya C okal, Janosch Haber, and Massimo Poesio. 2023. Modeling brain representations of words' concreteness in context using gpt-2 and human ratings. Cognitive Science, 47(12):e13388

    work page 2023

  6. [6]

    o lte, and Andrea B \

    Marc Brysbaert, Matthias Buchmeier, Markus Conrad, Arthur M Jacobs, Jens B \"o lte, and Andrea B \"o hl. 2011. The word frequency effect. Experimental psychology

    work page 2011

  7. [7]

    Marc Brysbaert, Pawe Mandera, and Emmanuel Keuleers. 2018. The word frequency effect in word processing: An updated review. Current directions in psychological science, 27(1):45--50

    work page 2018

  8. [8]

    Marc Brysbaert, Amy Beth Warriner, and Victor Kuperman. 2014. Concreteness ratings for 40 thousand generally known english word lemmas. Behavior research methods, 46:904--911

    2014

Show all 42 references
  1. [9]

    Madalina Bucur and Costanza Papagno. 2021. An ale meta-analytical review of the neural correlates of abstract and concrete words. Scientific reports, 11(1):15727

    2021

  2. [10]

    Branden Chan, Stefan Schweter, and Timo M \"o ller. 2020. https://doi.org/10.18653/v1/2020.coling-main.598 G erman`s next language model . In Proceedings of the 28th International Conference on Computational Linguistics, pages 6788--6796, Barcelona, Spain (Online). Internation...

    2020 doi

  3. [11]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. https://doi.org/10.18653/v1/N19-1423 BERT : Pre-training of deep bidirectional transformers for language understanding . In Proceedings of the 2019 Conference of the North A merican Chapter of the Associat...

    2019 doi

  4. [12]

    Birk Diedenhofen and Jochen Musch. 2015. cocor: A comprehensive solution for the statistical comparison of correlations. PloS one, 10(4):e0121945

    2015

  5. [13]

    Christian J Fiebach and Angela D Friederici. 2004. Processing concrete words: fmri evidence against a specific right-hemisphere involvement. Neuropsychologia, 42(1):62--70

    2004

  6. [14]

    Klaus Fliessbach, Susanne Weis, Peter Klaver, Christian Erich Elger, and Bernd Weber. 2006. The effect of word concreteness on recognition memory. NeuroImage, 32(3):1413--1421

    2006

  7. [15]

    Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. 2018. Learning word vectors for 157 languages. In Proceedings of the International Conference on Language Resources and Evaluation (LREC 2018)

    2018

  8. [16]

    Martin N Hebart, Charles Y Zheng, Francisco Pereira, and Chris I Baker. 2020. Revealing the multidimensional mental representations of natural objects underlying human similarity judgements. Nature human behaviour, 4(11):1173--1185

    2020

  9. [17]

    Hsu-Wen Huang and Kara D Federmeier. 2015. Imaginative language: What event-related potentials have revealed about the nature and source of concreteness effects. Language and linguistics, 16(4):503--515

    2015

  10. [18]

    Carlton T James. 1975. The role of semantic information in lexical decisions. Journal of Experimental Psychology: Human Perception and Performance, 1(2):130

    1975

  11. [19]

    Philipp Kanske and Sonja A Kotz. 2010. Leipzig affective norms for german: A reliability study. Behavior research methods, 42:987--991

    2010

  12. [20]

    Maximilian K \"o per and Sabine Schulte Im Walde. 2016. Automatically generated affective norms of abstractness, arousal, imageability and valence for 350 000 german lemmas. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pa...

    2016

  13. [21]

    Nikolaus Kriegeskorte and Rogier A Kievit. 2013. Representational geometry: integrating cognition, computation, and the brain. Trends in cognitive sciences, 17(8):401--412

    2013

  14. [22]

    Nikolaus Kriegeskorte, Marieke Mur, and Peter A Bandettini. 2008. Representational similarity analysis-connecting the branches of systems neuroscience. Frontiers in systems neuroscience, 2:249

    2008

  15. [23]

    Alessandro Lenci, Magnus Sahlgren, Patrick Jeuniaux, Amaru Cuba Gyllensten, and Martina Miliani. 2022. A comparative evaluation and analysis of three generations of distributional semantic models. Language resources and evaluation, 56(4):1269--1313

    2022

  16. [24]

    Guido L \"o hr. 2024. Does the mind care about whether a word is abstract or concrete? why concreteness is probably not a natural kind. Mind & Language, 39(5):627--646

    2024

  17. [25]

    Gonzalo Mart \' nez, Juan Diego Molero, Sandra Gonz \'a lez, Javier Conde, Marc Brysbaert, and Pedro Reviriego. 2025. Using large language models to estimate features of multi-word expressions: Concreteness, valence, arousal. Behavior Research Methods, 57(1):1--11

    2025

  18. [26]

    Maria Montefinese. 2019. Semantic representation of abstract and concrete words: A minireview of neural evidence. Journal of neurophysiology, 121(5):1585--1587

    2019

  19. [27]

    Andreas Müller . 2015. https://devmount.github.io/GermanWordEmbeddings Analyse von Wort-Vektoren deutscher Textkorpora

    2015

  20. [28]

    Subba Reddy Oota, Emin C elik, Fatma Deniz, and Mariya Toneva. 2024. https://doi.org/10.18653/v1/2024.acl-long.462 Speech language models lack important brain-relevant semantics . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volum...

    2024 doi

  21. [29]

    Allan Paivio, John C Yuille, and Stephen A Madigan. 1968. Concreteness, imagery, and meaningfulness values for 925 nouns. Journal of experimental psychology, 76(1p2):1

    1968

  22. [30]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-learn: Machine learning in P ython. Journal of Machine L...

    2011

  23. [31]

    Francisco Pereira, Bin Lou, Brianna Pritchett, Samuel Ritter, Samuel J Gershman, Nancy Kanwisher, Matthew Botvinick, and Evelina Fedorenko. 2018. Toward a universal decoder of linguistic meaning from brain activation. Nature communications, 9(1):963

    2018

  24. [32]

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9

    2019

  25. [33]

    Jamie Reilly, Cory Shain, Valentina Borghesani, Philipp Kuhnke, Gabriella Vigliocco, Jonathan E Peelle, Bradford Z Mahon, Laurel J Buxbaum, Asifa Majid, Marc Brysbaert, et al. 2024. What we mean when we say semantic: Toward a multidisciplinary semantic glossary. Psychonomic bu...

    2024

  26. [34]

    Valery Solovyev. 2020. Concreteness/abstractness concept: State of the art. In International Conference on Cognitive Sciences, pages 275--283. Springer

    2020

  27. [35]

    Ilia Sucholutsky, Lukas Muttenthaler, Adrian Weller, Andi Peng, Andreea Bobu, Been Kim, Bradley C Love, Erin Grant, Iris Groen, Jascha Achterberg, et al. 2023. Getting aligned on representational alignment. arXiv preprint arXiv:2310.13018

    2023 arXiv

  28. [36]

    Bill Thompson and Gary Lupyan. 2018. Automatic estimation of lexical concreteness in 77 languages. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 40

    2018

  29. [37]

    Jacopo Turini and Melissa Le-Hoa V \ o . 2022. Hierarchical organization of objects in scenes is reflected in mental representations of objects. Scientific Reports, 12(1):20068

    2022

  30. [38]

    Raphael Vallat. 2018. Pingouin: statistics in python. J. Open Source Softw., 3(31):1026

    2018

  31. [39]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems, 30

    2017

  32. [40]

    Christian Wartena. 2024. Estimating word concreteness from contextualized embeddings. In Proceedings of the 20th Conference on Natural Language Processing (KONVENS 2024), pages 81--88

    2024

  33. [41]

    Evan J Williams. 1959. The comparison of regression variables. Journal of the Royal Statistical Society: Series B (Methodological), 21(2):396--399

    1959

  34. [42]

    Tal Yarkoni, David Balota, and Melvin Yap. 2008. Moving beyond coltheart’s n: A new measure of orthographic similarity. Psychonomic bulletin & review, 15(5):971--979

    2008

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.